🧱 Smart Tile SEGS

If you've used Impact Pack's DetailerForEach, you know the SEGS format: a bundle of detected regions, each carrying its own cropped image, mask, and metadata. The Smart Tile SEGS node is the trick that borrows that whole machinery for upscaling - it converts your tile grid into SEGS, one SEG per tile, so every tool that knows how to chew on SEGS (DetailerForEach, SEGSLabelAssign, the whole Impact Pack ecosystem) suddenly knows how to process your tiles. It's the bridge between the calculator and the detailer in the "Smart Tile" subsystem of ComfyUI-ArchAi3d-Qwen (Amir Ferdos's pack).

Why it exists

Impact Pack already has a MakeTileSEGS, but it only supports square tiles and computes its own grid internally. This node is deliberately different: it takes explicit rectangular tile dimensions and a grid count from the calculator, and honors them. The source says it plainly - explicit grid, rectangular tiles, no guessing. That's what makes the calculator → SEGS handoff exact rather than approximate.

Inputs

• image - the upscaled image from the calculator.
• tile_width / tile_height / tiles_x / tiles_y / tile_padding - the grid, straight from the calculator's outputs.
• bundle (optional) - connect the calculator's bundle and it overrides all of the above. The intended wiring.
• filter_in_segs_opt / filter_out_segs_opt - SEGS filters. Only include tiles that overlap with these SEGS (e.g., a face-detection pass - detail only the tiles containing faces), or exclude tiles that overlap. This is the feature that makes "detail the faces, leave the rest alone" a one-cable job.
• crop_factor - extra context around each tile for better blending (default 1.5).

Outputs: segs (the SEGS, ready for DetailerForEach or the pack's own Detailer/Sampler) and bundle (pass-through).

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen
pip install -r requirements.txt

Or ComfyUI Manager → "ArchAi3d Qwen". No model downloads. Note it's SEGS, so you need Impact Pack installed to get the most out of the downstream nodes - the pack references Impact Pack's DetailerForEach by name.

The honest take

This is the node that makes the whole Smart Tile line make sense: it's Impact Pack's tiling behavior, but with the calculator's precision driving it. The filter_in_segs_opt input is genuinely worth your time - "only redetail the tiles that overlap a face detection" is a real workflow, and few tiled upscalers make it this easy. If you're new to SEGS, one warning: SEGS is a tuple type with its own convention (image size + a list of segments), so if a node says it wants SEGS, it means this SEGS - and Impact Pack's SEGS nodes are the natural next stop.